Marvin Minsky (1977), MIT Artificial Intelligence Laboratory, A.I. Memo No. 430, June 1977, 23 pp. — https://hdl.handle.net/1721.1/5763. Also an invited paper in Proceedings of the Fifth International Joint Conference on Artificial Intelligence (IJCAI-77), Cambridge, Mass., August 1977, vol. 2, pp. 1083–1092.
Key takeaways
- The mind is “an organized society of intercommunicating ‘agents’”, each “by itself, very simple”. The opening example is a child’s play with blocks, in which WRECKER, BUILDER, PUT and GRASP serve PLAY-WITH-BLOCKS, and PLAY loses out to I’M-GETTING-HUNGRY.
- Agents too simple for language communicate by sending nothing: “Each of an agent’s data sources is a FIXED location in (short term) memory.” In effect these are global variables, and “the agents are the STM”.
- Connectivity follows a specificity gradient: few, brain-wide top-level channels, and progressively local sub-societies that can talk only through the levels above them.
- Persistence-memory and a transient case-shift let a young mind survive interruptions without a recursive stack. Recursion in adults is “probably an illusion” of description (Note 7).
- The “cognitive cases” an infant’s agents use internally would later be “relatively easy to encode into external symbols”. On this view learning language is “learning to translate between languages”, and linguistic universals reflect early internal uniformities.
Standing in the record
Filed on 2026-10-01 at the owner’s request, as part of Minsky’s Society of Mind programme. Of the four memos filed in this pass, it is the first to state the society-of-agents view, and it names the theory “The Society of Minds” (Note 1). There it is described as “a theory being developed in collaboration with Seymour Papert”, which “we hope to publish … within the next year or so”. Whether this is the first published statement anywhere is the usual attribution, and I could not confirm it: Minsky’s AIM-516 credits a principle of the theory to Minsky and Papert’s 1974 University of Oregon lectures, which I did not see.
The paper says how it builds on the frame memo (LIT-410). It is “in part a sequel” to it. Its fixed data locations are “an extension of the ‘common terminal’ idea in my paper on frame-systems”. Its two-way memory addressing revisits that memo’s two-level frame matching. In turn, K-lines (LIT-402) says its own K→P connections correspond roughly to this paper’s c-lines, and offers itself as a complement. The book-length Society of Mind (1986), filed alongside by another pass, is the publication this paper promises.
NOTE-344 is the close reading of 2026-10-01, and it placed the work: Active. It is worth reading as the earliest mechanism-level statement of the programme. It is not evidence for it. The neurological parts (c-lines as white matter, the laminar hypothesis) are offered “with no pretense that there is any solid evidence”.
Neighbours in this record:
- Collective intelligence. The computational account of collective intelligence (LIT-046) asks what a collective computes that its members cannot. This paper runs the analogy the other way. A mind is a society, but one whose members are “all equally mindless”, so the social analogy is “poor” (Note 9).
- Group minds. Schwitzgebel’s argument that a nation could be conscious (LIT-159) needs collectives of minded members. Minsky’s society needs none.
- Extended mind. The extended-mind thesis (LIT-097) moves cognition outward from the head. This paper moves it downward into parts.
These pairings are mine. The paper carries no instruction for machine-learning practice, and no anthology topic holds it.